Simulation and Research Wind-Solar-Hydrogen Energy Storage Microgrid Based on DQN Deep Reinforcement Learning
DQN深層強化学習に基づく風力・太陽光・水素エネルギー貯蔵マイクログリッドのシミュレーションと研究 (AI 翻訳)
Zhihao Zhang, Jiayan Xie, Huie Zhang, Jianqiao Sun, Yingchao Dong, Jiang-Sheng Wang
🤖 gxceed AI 要約
日本語
本論文は、新疆の風力・太陽光の出力抑制問題を解決するため、風力・太陽光・水素貯蔵ハイブリッドマイクログリッドを構築し、DQN深層強化学習を用いた最適運用スケジューリングを提案。シミュレーションの結果、年間の風力・太陽光出力抑制率を4.1%に低減、総合エネルギー利用率を93.7%に向上、電源信頼性99.5%を達成し、運用コストを削減した。AIベースのスケジューリングが従来手法より優れていることを示した。
English
This paper constructs a wind-solar-hydrogen hybrid microgrid for Xinjiang, China, using DQN deep reinforcement learning for intelligent scheduling. The simulation results show that the proposed AI-based optimal scheduling can reduce annual wind and solar curtailment to 4.1%, increase comprehensive energy utilization to 93.7%, and ensure power supply reliability of 99.5% while lowering operating costs. The approach outperforms traditional and PPO scheduling methods.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
新疆の事例だが、日本の離島や山間部などの再生可能エネルギー導入促進と水素貯蔵の組み合わせに示唆を与える。日本でも風力・太陽光の出力抑制問題が発生しており、AIを用いた最適制御の有効性を実証している点が重要。
In the global GX context
Although focused on Xinjiang, China, this paper provides insights for regions with high renewable penetration and curtailment issues, such as remote areas. The use of DQN for microgrid scheduling demonstrates technical and economic viability for zero-carbon power supply, relevant to global energy transition efforts.
👥 読者別の含意
🔬研究者:Researchers in microgrid optimization and reinforcement learning for energy systems can use this as a benchmark for DRL-based scheduling.
🏢実務担当者:Energy engineers and system operators can consider DQN-based scheduling for reducing curtailment and improving efficiency in renewable-integrated microgrids.
🏛政策担当者:Policymakers interested in green hydrogen and remote area electrification can note the technical feasibility and cost benefits demonstrated.
📄 Abstract（原文）
In response to Xinjiang's high "wind and solar curtailment" rate due to abundant new energy but limited grid capacity and low power-supply reliability in rural areas, this paper constructs a wind-solar-hydrogen-storage hybrid microgrid system according to the region's climate and load characteristics. The model integrates wind turbines, photovoltaic arrays, PEM electrolyzers, hydrogen storage tanks, solid oxide fuel cells, battery energy storage and representative rural loads. It innovatively uses the deep reinforcement learning (DQN) algorithm for intelligent scheduling, with state variables covering renewable output, forecast deviation, battery SOC, hydrogen tank pressure and load demand, and with coordinated actions for hydrogen production, fuel-cell generation and battery charging or discharging. A reward function with reliability priority, renewable-energy consumption penalty and operating-cost penalty is designed to guide operation under surplus, deficit and fluctuating conditions. Entity modeling and multi-scenario simulation are carried out on the MATLAB/Simulink platform. Simulation results show that compared with traditional, fixed-threshold and PPO scheduling, AI-based optimal scheduling can effectively reduce the annual wind and solar curtailment rate to 4.1%, increase the comprehensive energy utilization rate to 93.7%, ensure power supply reliability of 99.5%, and lower annual operating costs. Supplementary disturbance tests further indicate that the learned policy remains feasible under renewable-output attenuation, load-surge and forecast-error conditions. This study validates the system's technical feasibility and economic advantages under regional operating conditions and provides engineering support for Xinjiang's green hydrogen industry chain and zero-carbon power supply in remote areas.
🔗 Provenance — このレコードを発見したソース
- semanticscholar https://doi.org/10.1109/epsic70071.2026.11590705first seen 2026-07-13 07:03:34
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。